Rootless containers as self-hosted runners
Faster, cheaper CI/CD and an AI agent playground
Cloud runners and LLM APIs each run their own billing meter. Use both together
and you are billed twice. This talk shows how we use self-hosted openSUSE
containers for large codebases, and how we re-used it to eliminate that overlap.
We walk through our Linux kernel testing pipeline: cache management,
cross-architecture builds, checkpatch per-commit, and more.
Then, we explore how we get the most of LLMs by running them inside the same
container. Using pi.dev as the agent harness, we can swap the model, and tune
the system prompt for one-shot requests. The agent runs after the top-level
pipeline completes, with CI annotations (warnings and errors), being part of
the prompt. The agent is instructed to assess the issues, and validate its
claims by rebuilding and testing. As the output, it generates git patches, and
summarizes the session into a structured review summary.
Licensed to the public under https://creativecommons.org/licenses/by-sa/4.0/